AI Engineers: Building Minds, Breaking Theirs

Author: Kristoffer Dave Tabong | July 22, 2025

An honest glimpse into what it really takes to build intelligent machines.

1. AI Engineering is 80% Math, 10% Debugging, 10% Existential Crisis
   At first, AI engineering seems like a technical pursuit—just data, models, and code. But somewhere between calculating loss functions and fixing shape mismatches, you find yourself staring into the void, wondering if you're simply a biological neural net teaching a digital one to replace you. It’s not just engineering—it’s philosophy, panic, and purpose rolled into one.
2. We gave Machines Brains but forgot to teach them Common Sense
   We’ve built machines that can defeat world champions in chess and solve complex problems faster than any human. And yet, those same systems can mistake a chihuahua for a blueberry muffin. The gap between raw intelligence and basic common sense remains wide and oddly hilarious. It’s progress, but not without glitches that force us to rethink how we define “smart.”
3. Data is the New Oil... and AI Engineers are the Janitors
   Everyone loves to say that data is the new oil, but they rarely mention who does the refining. Before any AI model can learn, someone has to clean the data; scrubbing out typos, fixing inconsistencies, handling nulls, and deciphering cryptic columns. That someone is the AI engineer, armed with Python libraries, regular expressions, and relentless persistence. It's not glamorous, but it's absolutely essential.
4. Half the Job is convincing Stakeholders that AI isn’t Magic
   Explaining AI to non-technical stakeholders often feels like a stand-up routine no one realizes is a tragedy. The boss asks if the model can just “learn on its own,” as if AI were a toddler raised by YouTube. You try to explain that training is trial and error, not magic. Eventually, you give in and say “Yes, it’s AI-powered”—but inside, you die a little.
5. Overfitting: When your Model aces the test and fails in Life
   In training, your model performs flawlessly—predicting with pinpoint accuracy. But once it hits the real world, it falls apart, recommending cat food to someone shopping for dog shampoo. This is overfitting in action: a model that memorized the past so well it can’t handle anything new. It’s a painful reminder that good grades in the lab don’t guarantee success in production.
6. AI Bias: The machine learns from our mistakes—then scales them
   When you feed AI historical data, it doesn’t just learn patterns—it absorbs all the human flaws baked into them. What starts as a quest for efficiency can end in a system that reinforces societal bias with mathematical confidence. Bias in AI isn’t just a coding issue—it’s a mirror, reflecting and amplifying our collective failures.
7. GPT Engineers: Teaching a Giant Parrot to Sound Smart
   Working with large language models like GPT feels less like programming and more like coaxing a very articulate parrot. These models don’t understand—they autocomplete. They lack emotion, intention, or awareness, yet produce answers with uncanny fluency. They can’t feel your pain, but they’ll describe it better than you ever could, as long as the prompt is clear and the Wi-Fi’s stable.
8. AI Models Use More Energy Than a Small Town
   That quick chatbot reply? It comes at a cost—massive server farms powering every token it generates. Training and running large models consume more electricity than entire towns. Each prompt is backed by energy-intensive computation, making us quietly wonder if asking a model to write haikus is worth the environmental bill.
9. The secret no one wants to admit: We don’t know why it works
   Behind the confident dashboards and technical papers, there’s a truth many engineers whisper only in private: we don’t fully understand how some of these models do what they do. We label it “emergent behavior”, which sounds impressive but mostly means “we’re guessing.” Despite all our efforts at explainable AI, mystery still lurks at the core.